Keywords
partial least squares (pls), neural net pls (nnpls), kernel pls (kpls), on-line process monitoring and control, biological wastewater treatment plants, supervisory control
Start Date
1-7-2006 12:00 AM
Abstract
Despite of promising results in research, advanced control strategies fail to gain trust in wastewater treatment practice. Due to the sensitivity of the biological processes to disturbances, operators are often unable to find the causes of faults due to the lack of effective real-time on-line monitoring. Strategies for on-line monitoring are therefore essential to enhance biological process control. Therefore, a suitable multivariate soft-sensor is desired for fault detection and control for a pilot-scale sequencing batch reactor (SBR) system to allow effluent quality to be estimated well before off-line analysis is finished. For this purpose, several multivariate methods are available, including (linear) partial least squares (PLS), Neural Net PLS (NNPLS) and Kernel PLS (KPLS). While non-linear extensions of PLS such as NNPLS require fitting of non-linear functions, KPLS does not. KPLS is based on a non-linear transformation of the process data, followed by the fitting of a linear PLS model between the transformed inputs and outputs. PLS, NNPLS and KPLS were compared regarding their ability to predict effluent quality data and their computational requirements. While (linear) PLS and NNPLS lead to acceptable prediction, KPLS results in poor model performance. Moreover, the computational requirement of KPLS were large compared to PLS and NNPLS. When comparing PLS and NNPLS to each other, it was found that NNPLS leads to the best possible prediction given the experimental data set, while the extra computational requirements are minimal.
Comparison of linear and non-linear PLS methods for soft-sensing of an SBR for nutrient removal
Despite of promising results in research, advanced control strategies fail to gain trust in wastewater treatment practice. Due to the sensitivity of the biological processes to disturbances, operators are often unable to find the causes of faults due to the lack of effective real-time on-line monitoring. Strategies for on-line monitoring are therefore essential to enhance biological process control. Therefore, a suitable multivariate soft-sensor is desired for fault detection and control for a pilot-scale sequencing batch reactor (SBR) system to allow effluent quality to be estimated well before off-line analysis is finished. For this purpose, several multivariate methods are available, including (linear) partial least squares (PLS), Neural Net PLS (NNPLS) and Kernel PLS (KPLS). While non-linear extensions of PLS such as NNPLS require fitting of non-linear functions, KPLS does not. KPLS is based on a non-linear transformation of the process data, followed by the fitting of a linear PLS model between the transformed inputs and outputs. PLS, NNPLS and KPLS were compared regarding their ability to predict effluent quality data and their computational requirements. While (linear) PLS and NNPLS lead to acceptable prediction, KPLS results in poor model performance. Moreover, the computational requirement of KPLS were large compared to PLS and NNPLS. When comparing PLS and NNPLS to each other, it was found that NNPLS leads to the best possible prediction given the experimental data set, while the extra computational requirements are minimal.
